Bug 3240 – std.numeric.findRoot only works with real

Trying to use std.numeric.findRoot with other types than real gives template instantiation errors originating from lines 599 and 560 in numeric.d (DMD 2.031, SVN rev. 1200): 599 if (a==0) c = ieeeMean(copysign(0.0L, b), b); 560 else if (b==0) c = ieeeMean(copysign(0.0L, a), a); If a and b are double, I get the following errors: /usr/local/include/d/phobos2/std/numeric.d(599): Error: template std.math.ieeeMean(T) does not match any function template declaration /usr/local/include/d/phobos2/std/numeric.d(599): Error: template std.math.ieeeMean(T) cannot deduce template function from argument types !()(real,double) /usr/local/include/d/phobos2/std/numeric.d(600): Error: template std.math.ieeeMean(T) does not match any function template declaration /usr/local/include/d/phobos2/std/numeric.d(600): Error: template std.math.ieeeMean(T) cannot deduce template function from argument types !()(real,double) This is because IFTI doesn't work with implicit conversions (bug 1641 and probably others), and std.math.copysign() always returns a real.

Um.. like the actual pasted error messages say, the line numbers should be 599 and 600, not 500 and 560. Sorry, don't know how that happened.

Good catch. That's caused by a change to the signature of ieeeMean(). Trivially fixable by inserting casts, though arguably copySign should become a templated function. BTW I wrote this code for Tango, it will eventually have a signature change to match the new Phobos style. It predates ranges, for example, even though it's a perfect example of a floating-point range-based algorithm. Thus, it's not as generic as it could be. Don.

There's an even simpler fix -- i.e. less typing. :) Since both a, b and c are of type T, this also works: 599 if (a==0) c = ieeeMean!T(copysign(0.0L, b), b); 600 else if (b==0) c = ieeeMean!T(copysign(0.0L, a), a); I've added this fix to my local Phobos. I just tried this for fun, and expected to get an error from trying to implicitly cast real to double. But for some reason it worked.

(In reply to comment #3) > There's an even simpler fix -- i.e. less typing. :) > > Since both a, b and c are of type T, this also works: > > 599 if (a==0) c = ieeeMean!T(copysign(0.0L, b), b); > 600 else if (b==0) c = ieeeMean!T(copysign(0.0L, a), a); > > I've added this fix to my local Phobos. I just tried this for fun, and expected > to get an error from trying to implicitly cast real to double. But for some > reason it worked. Don, sorry, I took this over before seeing you are considering working on it. You may want to take ownership. Thanks!

Fixed in SVN commit 1246.